feat: complete M7 cutover preparation

server-rs/crates/api-server/src/app.rs

@@ -6,8 +6,11 @@ use axum::{
     middleware,
     routing::{get, post},
 };
-use tower_http::trace::{DefaultOnFailure, DefaultOnRequest, DefaultOnResponse, TraceLayer};
-use tracing::{Level, info_span};
+use tower_http::{
+    classify::ServerErrorsFailureClass,
+    trace::{DefaultOnRequest, TraceLayer},
+};
+use tracing::{Level, Span, error, info, info_span, warn};
 
 use crate::{
     ai_tasks::{
@@ -86,6 +89,8 @@ use crate::{
 
 // 统一由这里构造 Axum 路由树，后续再逐项挂接中间件与业务路由。
 pub fn build_router(state: AppState) -> Router {
+    let slow_request_threshold_ms = state.config.slow_request_threshold_ms;
+
     Router::new()
         .route(
             "/healthz",
@@ -688,8 +693,39 @@ pub fn build_router(state: AppState) -> Router {
                     )
                 })
                 .on_request(DefaultOnRequest::new().level(Level::INFO))
-                .on_response(DefaultOnResponse::new().level(Level::INFO))
-                .on_failure(DefaultOnFailure::new().level(Level::ERROR)),
+                .on_response(move |response: &axum::response::Response, latency: std::time::Duration, span: &Span| {
+                    let latency_ms = latency.as_millis().min(u64::MAX as u128) as u64;
+                    let status = response.status().as_u16();
+                    let slow_request = latency_ms >= slow_request_threshold_ms;
+                    span.record("status", status);
+                    span.record("latency_ms", latency_ms);
+                    if slow_request {
+                        warn!(
+                            parent: span,
+                            status,
+                            latency_ms,
+                            slow_request = true,
+                            "http request completed slowly"
+                        );
+                    } else {
+                        info!(
+                            parent: span,
+                            status,
+                            latency_ms,
+                            slow_request = false,
+                            "http request completed"
+                        );
+                    }
+                })
+                .on_failure(|failure: ServerErrorsFailureClass, latency: std::time::Duration, span: &Span| {
+                    let latency_ms = latency.as_millis().min(u64::MAX as u128) as u64;
+                    error!(
+                        parent: span,
+                        latency_ms,
+                        failure = %failure,
+                        "http request failed"
+                    );
+                }),
         )
         // request_id 中间件先进入请求链，确保后续 tracing、错误处理和响应头层都能复用同一份请求标识。
         .layer(middleware::from_fn(attach_request_context))

server-rs/crates/api-server/src/config.rs

@@ -54,6 +54,7 @@ pub struct AppConfig {
     pub llm_request_timeout_ms: u64,
     pub llm_max_retries: u32,
     pub llm_retry_backoff_ms: u64,
+    pub slow_request_threshold_ms: u64,
 }
 
 impl Default for AppConfig {
@@ -104,6 +105,7 @@ impl Default for AppConfig {
             llm_request_timeout_ms: DEFAULT_REQUEST_TIMEOUT_MS,
             llm_max_retries: DEFAULT_MAX_RETRIES,
             llm_retry_backoff_ms: DEFAULT_RETRY_BACKOFF_MS,
+            slow_request_threshold_ms: 1_000,
         }
     }
 }
@@ -305,6 +307,12 @@ impl AppConfig {
             config.llm_retry_backoff_ms = llm_retry_backoff_ms;
         }
 
+        if let Some(slow_request_threshold_ms) =
+            read_first_positive_u64_env(&["GENARRATIVE_SLOW_REQUEST_THRESHOLD_MS"])
+        {
+            config.slow_request_threshold_ms = slow_request_threshold_ms;
+        }
+
         config
     }